tag:blogger.com,1999:blog-7729093380675162051.post4976432077726657451..comments2024-03-28T13:56:47.604+11:00Comments on moyhu: Ryan's codeNick Stokeshttp://www.blogger.com/profile/06377413236983002873noreply@blogger.comBlogger14125tag:blogger.com,1999:blog-7729093380675162051.post-82116555724765778492011-02-17T09:49:34.969+11:002011-02-17T09:49:34.969+11:00The other code is apparently something he was work...The other code is apparently something he was working on after the acceptance of the paper. I too encounter an error during the validation process with the AVHRR data, I assumed (perhaps erroneously) that it was innocuous.<br /><br />In case Ryan see this, I'll repeat my recommendations:<br /><br /><i>My recommendation would be to make code like this slightly less turn-key: Give a list of instructions for how to run it, and an explanation for how to verify each step was performed successfully.</i>Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-21924748563561282722011-02-16T23:24:31.325+11:002011-02-16T23:24:31.325+11:00Carrick,
It ran to completion in that .txt version...Carrick,<br />It ran to completion in that .txt version - took 46 minutes on my 4 Gb Windows setup. Now I can try to see what it does.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-71408477994618442912011-02-16T17:52:06.798+11:002011-02-16T17:52:06.798+11:00Ah, that may be my problem. The RO10_Script.R vers...Ah, that may be my problem. The RO10_Script.R version (which I used) is different to the RO10 Code.txt version, and the line I'm stuck on is replaced by a line:<br />D = sqrt(colSums(dat$X ^ 2) / arg$n.eff)<br />which should work. OK, I'll try that. Thanks.<br /><br />Still, it's a puzzle. I found in my original version of the .R file the same error as noted <a href="http://climateaudit.org/2011/02/07/eric-steigs-trick/#comment-254004" rel="nofollow">here</a>. Ryan responded, saying that he'd put up a fixed version, and indeed, the .R code had been fixed. <br /><br />But the actual name of the R script used is not mentioned in that thread.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-30520277658704840872011-02-16T14:23:16.832+11:002011-02-16T14:23:16.832+11:00And the answer is "yes". It definitely ...And the answer is "yes". It definitely gets called... a bunch of times.<br /><br />Note, I'm using the file "RO10 Code.txt" as per the readme file.Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-23147330789934485242011-02-15T17:37:08.485+11:002011-02-15T17:37:08.485+11:00Thanks, CarrickThanks, CarrickNick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-77345505948787695662011-02-15T15:05:06.054+11:002011-02-15T15:05:06.054+11:00I will try that tomorrow and let you know.I will try that tomorrow and let you know.Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-82947978709889753652011-02-15T14:17:27.836+11:002011-02-15T14:17:27.836+11:00One of the things I'm curious about is timing....One of the things I'm curious about is timing. I spent a lot of time tweaking TempLS. R is good if you keep the numerical work in big chunks.<br /><br />Carrick, if you ever run the code again, I'd be really interested to know if it goes into that tsvdRegFn() function.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-76833252645751276302011-02-15T08:42:36.506+11:002011-02-15T08:42:36.506+11:00Deepclimate, I would think, properly written, the ...Deepclimate, I would think, properly written, the codes would run in about the same amount of tine on either platform. <br /><br />I think the biggest issue, as a developer, is which environment is faster to generate a final product in. In my experience these days, for most code, more time is spent in developing and in the code verification stage than actually running the final product.Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-55198725652526141552011-02-15T08:14:47.511+11:002011-02-15T08:14:47.511+11:00Would not implementing RegEM in R slow it down fro...Would not implementing RegEM in R slow it down from the original Matlab (interpreted vs compiled)?<br /><br />Or are the operations "standard" enough that library stats processing calls can be combined in such away as to avoid a performance penalty?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-67614148928615355062011-02-15T07:08:37.875+11:002011-02-15T07:08:37.875+11:00Carrick,
Yes, I've restored the comment that g...Carrick,<br />Yes, I've restored the comment that got caught in the spam filter.<br /><br />I ran the code with original tolerance settings, but still got the same error sequence.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-16284345393021007032011-02-15T06:46:25.891+11:002011-02-15T06:46:25.891+11:00Nick, I'm looking forward to your review of Ry...Nick, I'm looking forward to your review of Ryan's code.<br /><br />One thing I noticed was it doesn't terminate if it can't download all of the data files (you can a "cannot open connection" error, and it keeps running, generating a long trail of error messages).<br /><br />My recommendation would be to make code like this slightly less turn-key: Give a list of instructions for how to run it, and an explanation for how to verify each step was performed successfully.<br /><br />Also, on speed, my code ran in about an hour on a rather horsey computer. I'd love to see what Ryan is running his on, if his only took 17 minutes.<br /><br />I have a couple of comments in the "On words" thread that appear to be in limbo. THere are two versions, if possible could you ixnay the older version, and approve the more recent version?Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-9502254765779330152011-02-15T06:17:34.918+11:002011-02-15T06:17:34.918+11:00KC,
If you click on that trashcan symbol, there...KC,<br />If you click on that trashcan symbol, there's an option to obliterate comments compleetely.<br /><br />Anon,<br />That's my mistake - my original fix was not to remove the arg from the call, but to add it to the definition. It wasn't there in the original.<br /><br />Carrick,<br />That's very interesting, and for me encouraging. Ryan's response was that I shouldn't be in a tsvd part of the code. I think the bit of code I am in is not right - I don't see how it makes sense to set div$C to NA and then make use of it. But it is indeed possible that I shouldn't be there. <br /><br />I did modify the tolerances to speed up the run, since I was only interested at this stage in getting the program sequence to run. I'll try again with the original tolerances.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-42907349485477125392011-02-15T04:04:17.890+11:002011-02-15T04:04:17.890+11:00I'll note the code ran for me (version 2.12.1,...I'll note the code ran for me (version 2.12.1, 64 bit code, Mac OS 10.6.6).<br /><br />How sure are you this isn't a memory related problem?Carrickhttps://www.blogger.com/profile/03476050886656768837noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-83246527940976380122011-02-15T00:49:01.748+11:002011-02-15T00:49:01.748+11:00But following your own link to getEigen, the funct...But following your own link to getEigen, the function definition looks like this:<br /><br />getEigen = function(Y, C = NA, dat = all.avhrr, pc.data = NA, DoFL = 12, form = 11, idx = NA, pc.max = 28, maxiter = 5000, tol = 0.0005,<br />n.eof = 9, type = "tsvd", wt = TRUE, progressive = c(T, 1, T, 1), alpha = 10, dat.st = 1982, dat.en = c(2006, 12), covr = FALSE) {<br /><br />covr is there as the last (presumably optional) argument?Anonymousnoreply@blogger.com